LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
A market maker is the counterparty who quotes a bid and an ask so someone else can trade. In a toy market where the "share" is a die, the center of that quote is the expected face, 3.5, and the expected gain per trade equals half the spread when volume is balanced. An option replaces the die with a discounted risk-neutral average from a calibrated model. That average is only as good as the model, and hedging costs eat the spread you hoped to keep.
Why it matters
The bid is where the public sells to the maker. The ask, or offer, is where the public buys. Without a positive gap, a trader could buy the low quote and sell the high one straight back. The lecture's dice desk makes the job concrete. A trader buys 1,000 shares at $3. The maker is short those 1,000. The die shows 6. The trader's gain is $3 a share, $3,000. The maker's job is to choose the $3, not to guess the face.
If quotes sit around a sound center, the lecture's wealth sketch drifts up across random outcomes. If the center is poor, the sketch drifts down. You do not pick the path. You pick the quote.
The idea in plain words
The mid is defined as the guess, called theta, that minimizes mean squared error against the die. The error surface is a parabola. Its bottom is 3.5, the same number you get by averaging the six faces. Calculus makes the link formal: differentiate, set the derivative to zero, and the best theta is the expectation. A fair die's probabilities never change, so repeated rolls pull the running mean to 3.5.
Quote a bid and an ask around that mid. Assume equal volume on each side and apply the law of total expectation. Selling at the ask, gain is ask minus the face. Buying at the bid, gain is the face minus the bid. The expected gain per trade simplifies to half the spread.
Widening the spread raises that half-spread on paper and kills the trades in practice. One simulation with a very wide quote produced only two transactions. The real tradeoff is a hump, too tight on one side and too wide on the other. Illiquid markets quote wider. Liquid ones, helped by automated quoting, quote tight, and the tighter book in the lecture showed many more trades. The maker is still buying below the mid and selling above it, and is still exposed, just to different risks than a directional bet.
How to use it
An option does not settle when the trade prints. The payoff arrives at a maturity T, and you do not know the path. The fundamental theorem of asset pricing says today's fair value is a discounted risk-neutral expectation, an average in the same role as 3.5. You still choose a model: geometric Brownian motion, Heston, rough Bergomi, or another. Equity options in the lecture lean toward a stochastic-volatility model such as Heston.
Parameters are not a personal forecast. Calibrate them to the market's implied-volatility surface, the liquid instruments, and treat that fit as the risk-neutral measure you will use. Simulate paths, take the payoff at T, discount, and average. With the parameters held fixed, the average settles on that model's theoretical price. Around it you still widen for the spread you want, and for counterparty and liquidity adjustments.
What usually goes wrong
Convergence is to the model's price. A poor model can show a smooth theoretical average and a realized path of losses. The second catch is time. The die resolves at once. The option does not, so an unhedged book waits until T, and an imperfect hedge gives back spread. The more effective the hedge, the more of the theoretical half-spread a liquidity provider keeps, inside the lecture's accounting. Frictions and transaction costs decide how much survives.
Questions people ask
Why is the dice mid 3.5 if a die cannot show it?
It is the mean-squared-error minimizer and the average of the six faces. The quote is built around that average, not around a face you expect to roll.
What is expected gain per dice trade?
With equal volume on both sides, the lecture's calculation gives half the spread.
Does calibration make the option price true?
It makes the price consistent with the chosen model and the liquid surface. The average converges to that model's value. A misspecified model can still lose money after the quote.